Collision Prevention for Autonomous Drones

A dynamic collision risk monitoring model for unmanned aerial vehicles (UAVs) that uses track conformity to assess collision risk. The model calculates the three-dimensional deviation between a UAV's actual flight path and its planned route, and evaluates the deviation data to determine track conformity. This allows estimating collision risk based on how closely the UAV stays on its planned route. The model also considers factors like maneuvering delay time to provide recommended conflict resolution times.

Source

A method for collision avoidance in unmanned aerial vehicles (UAVs) using multivariate data fusion to improve accuracy and reliability compared to single-variable algorithms. The method involves fusing multidimensional monitoring data like GPS, radar, and optics through neural networks to perform weighted processing and classification. This allows complex collision avoidance warnings based on dynamic programming iteration of the fused data. The neural network extracts features, connects them, and sends output to a classifier. By fusing and weighting diverse data, it provides more comprehensive and robust collision avoidance compared to simplistic single-variable approaches.

Source

Safe path planning for swarms of unmanned aerial vehicles (UAVs) in dynamic and uncertain environments using reinforcement learning. The method involves modeling the path planning problem as a partially observable Markov decision process (POMDP) and applying a multi-agent deep reinforcement learning algorithm called MASAC to train the UAVs to autonomously navigate to targets while avoiding obstacles and maintaining formation. The algorithm uses a shared network for all UAVs to learn optimal joint actions.

Source

Collision avoidance method for multiple rotor drones flying in a 2D plane that improves the ability of drones to resolve collisions in view of low efficiency and stretched resource utilization in dense drone environments. The method involves calculating relative distances and speeds between drones, analyzing based on optimized route authority and priority rules, and selectively executing emergency collision avoidance, hovering, and heading maintenance maneuvers when collision conditions are met. The selective speed obstacle algorithm with new distance and priority rules improves collision avoidance success rate for dense drone clusters.

Source

Method for autonomous drones to safely navigate through uncertain and dynamic indoor environments by accurately perceiving collision risks. The method involves modeling the uncertainty in obstacle positions using random walks, constructing joint distributions for multiple obstacle positions, and quantifying collision risks using conditional Value at Risk (CVaR) instead of traditional risk metrics. By conditioning on the drone's path, CVaR captures tail risks and non-Gaussian distributions better. This allows finding optimal collision-free paths with the CVaR-A* algorithm while ensuring drone safety in dense obstacle environments.

Source

UAV collision avoidance method that balances safety and energy efficiency for drones operating in crowded airspace. It uses Markov Decision Processes (MDPs) to find optimal maneuvers to avoid collisions while minimizing energy consumption and safety costs. The method considers factors like UAV kinematics, environment uncertainty, and sensor errors. It relies on an MDP model to make real-time collision avoidance decisions that strike a balance between safety and energy efficiency.

Source

Self-organizing collision avoidance method for UAV swarms based on reward-modulated spiking neural networks. It allows decentralized, self-organizing, and local interaction strategies to implement self-organized collision avoidance technology to improve multi-UAV systems. The method involves constructing an information transfer network between drones, introducing brain-inspired decision rules, and implementing brain-inspired pattern formation mechanisms. The drones learn from local observations to make coordinated obstacle avoidance decisions based on the learned knowledge. The brain-inspired reward-modulated spiking neural network uses spike timing-dependent plasticity to learn from spike timing correlations.

Source

Real-time collision avoidance method for multiple drones where the collision points are unknown. It involves using onboard sensors and computer vision to detect potential collisions in real-time and dynamically adjust flight paths to avoid them. The method involves continuously monitoring the drone's surroundings and checking for obstacles in its flight path. If a potential collision is detected, the drone's flight plan is altered to avoid the obstacle. The algorithm does not rely on pre-determined collision points and can handle unexpected obstacles that suddenly appear in the drone's path.

Source

UAV collision avoidance system that uses a combination of motion parameters and environmental factors to more accurately predict and mitigate collision risks. The system analyzes the motion of both the UAV and potential collision objects, as well as factors like weather, lighting, and terrain, to provide a more comprehensive collision risk assessment. This allows the UAV to make better avoidance decisions in complex environments where other methods may be limited.

Source

Collision avoidance for multi-UAV formations using deep reinforcement learning. The method involves training a neural network to learn optimal collision avoidance behaviors for a group of UAVs forming a shape. The network takes as input the current UAV positions and velocities, as well as nearby obstacle information, and outputs desired accelerations for each UAV. The network is trained using a reinforcement learning algorithm to maximize a reward function that balances collision avoidance with maintaining the formation shape. This allows the UAVs to dynamically adapt and avoid collisions while maintaining the desired formation.

Source

Collision avoidance method and system for unmanned aerial vehicles (UAVs) operating in airspace shared with manned aircraft. The method involves using ADS-B data from manned aircraft and UAVs to detect potential collisions based on relative positions and speeds. If a collision risk is detected, the UAV performs collision avoidance maneuvers to separate from the manned aircraft. The system includes a collision risk determination unit and conflict relief unit. The collision logic uses the ADS-B data to classify collision scenarios and select appropriate relief strategies.

Source

Dynamic route planning system for autonomous drones that allows them to plan collision-free paths in complex environments without relying on external signals. The system uses reinforcement learning and task-oriented collision avoidance. It includes modules for environmental observation, judgment/evaluation, intelligent decision-making, action execution, task execution, and task solution. The learning involves Q-learning with neural networks to approximate the value function for generalizing to infinite states. The system can learn to autonomously plan routes, avoid obstacles, and complete tasks in dynamic, complex environments without GPS or remote control signals.

Source

Collaborative operation of multiple drones using a method that ensures safety and reliability when operating in groups. The method involves calculating stability metrics for each drone based on factors like overload ratio, wind speed, and historical flight data. It also calculates an aggregation degree based on proximity to other drones. Using these metrics, a neural network predicts safety probability. If below a threshold based on aggregation, a maintenance signal is sent. This helps prevent issues in one drone from cascading to others.

Source

Collaborative sensing and collision avoidance for multiple UAVs in dense airspace with limited communication resources. The method involves using onboard sensing to detect conflict risks between UAVs, quantifying the risk levels using a weighted network, and then finding optimal collision avoidance maneuvers that minimize communication and cost. UAVs share the maneuver sequences over limited links, allowing coordinated flight without centralized control. The method enables safe operation of dense UAV swarms where centralized planning is impractical due to bandwidth limitations.

Source

A method for preventing collisions between multiple monitored flight vehicles using augmented reality (AR) to accurately predict and warn against potential collisions. The method involves assigning virtual flight vehicles to each monitored flight vehicle, using their actual positions. Collision detection is done using this augmented fleet data. If a virtual collision is predicted, it's displayed on the monitored flight's control screen along with details on how to avoid it. This allows better collision prediction and response compared to using just real flight data.

Source

A swarm of drones that can dynamically adapt and collaborate in real-time to perform complex tasks by leveraging advanced computing, communication, and collaboration techniques. The system allows large numbers of drones to efficiently process and respond to changing needs and environments. It includes modules for communication, collaboration, collision avoidance, navigation, positioning, environmental adaptation, security, data processing, and real-time decision-making. The drones can dynamically update avoidance strategies to adapt to environmental changes.

Source

Reducing the difficulty of operating a drone while ensuring flight safety by using autonomous flight planning based on historical flight data. The method involves analyzing previous flights to generate flight plans that avoid obstacles and navigate around hazardous areas. It leverages the fact that drones often follow similar routes and encounter similar conditions, allowing the system to learn and reuse past flight data to create optimized autonomous flight plans. This reduces the need for manual piloting in repetitive or challenging scenarios.

Source

Efficient and reliable multi-UAV perception and avoidance using deep reinforcement learning. The method involves building a multi-UAV sensing and avoidance model based on deep reinforcement learning. It uses UAV status, target info, radar data, and nearest neighbor interaction to fuse sensor observations. This integrates radar and neighbor data for obstacle avoidance, improving perception beyond single sensors. The training involves stages: first, collision avoidance using only neighbors; second, with radar plus neighbors. This progressive training from simple to complex environments improves the model's reliability in heterogeneous environments.

Source

Preventing collisions between unmanned aerial vehicles (UAVs) and other aircraft using an onboard collision avoidance system. The system calculates evasion trajectories for the UAV based on real-time data received from nearby manned aircraft via ADS-B. If a potential collision is detected, the UAV can autonomously maneuver horizontally and vertically to avoid the other aircraft. If the pilot doesn't respond to collision warnings, the system can take control of the UAV to enforce the evasive maneuvers.

Source

Collision avoidance method for large-scale UAV formations that uses Markov decision-making and probability models to improve safety and reduce unnecessary alarms compared to traditional TCAS systems. The method considers factors like probabilistic tracking, uncertainty in sensor data, and relative speed/time to define formation separation areas. It also introduces a variable lambda to balance collision risk vs alarm cost. By modeling the encounter as an optimization problem with probabilistic dynamics, it finds the optimal collision avoidance logic for formation flight.

Source